Backups

[John Tulp]

On Fri, 2022-12-09 at 20:03 -0800, Doug Hardie wrote:
> > On Dec 4, 2022, at 1:42 PM, Doug Hardie <bc979 at lafn.org> wrote:
> > 
> > > On Dec 3, 2022, at 11:50 PM, Doug Hardie <bc979 at lafn.org> wrote:
> > > 
> > > I started to investigate using doveadm backup to backup my mail
> > > system.  I have a small number of users and the mail store is not
> > > large.  It uses maildir format.  I setup a test system that is not
> > > connected to the internet and started up dovecot.  I used the
> > > following command to backup one user:
> > > 
> > > doveadm backup -u ben remote:test
> > > 
> > > ben is the user is in the mail store.  Test is the actual server
> > > name.  That worked just fine.  The maildir was copied completely
> > > (as best as I can tell with ls).  Then I tried the second user:
> > > 
> > > doveadm backup -u jean remote:test
> > > 
> > > This gives 2 error messages:
> > > 
> > > doveadm(jean)[]<McRYOFI0jGN1MwAAmC+70w>: Error: Mailbox INBOX:
> > > Failed to get attribute
> > > vendor/vendor.dovecot/pvt/server/sieve/files/.dovecot: Mailbox
> > > attributes not enabled
> > > 
> > > doveadm(jean)[]<0IwxIlI0jGMgUwAAZU03Dg>: Error: Remote command
> > > returned error 65: ssh test doveadm dsync-server -ujean -U
> > > 
> > > In addition, the maildir directories are created, but there are no
> > > emails in any of them (e.g., cur).  What is the problem with the
> > > 2nd and why does it behave differently from the first?
> > 
> > I managed to resolve most of the issue.  I use pigeonhole on the
> > primary server.  Apparently not having pigeonhole installed on the
> > test machine caused the errors above.  The test machine was never
> > intended to receive mail hence, no need to install pigeonhole as the
> > LDA would never be used.  However, when the backup was running, it
> > choked on transferring the sieve file.  I have no idea where the
> > mentioned file resides as I couldn't find it anywhere on the primary
> > server.  Installing pigeonhole resolved the issue for all but one
> > user. With that user, I get the following error messages:
> > 
> > doveadm(doug)[]<NAfSLTASjWNWTAAAmC+70w>: Panic: file
> > istream-seekable.c: line 238 (read_from_buffer): assertion failed:
> > (*ret_r > 0)
> > Abort
> > 
> > doveadm(doug)[]<C2AvHjASjWN2gQAAZU03Dg>: Error: read(test) failed:
> > EOF (last sent=mail, last recv=mail_request (EOL))
> > 
> > doveadm(doug)[]<C2AvHjASjWN2gQAAZU03Dg>: Error: Remote command
> > returned error 134: ssh test doveadm dsync-server -udoug -U
> > 
> > In addition, only a few cur files are transferred in INBOX.
> >  Repeating the backup generates the same errors and no additional
> > emails are transferred.  I am wondering if the problem is something
> > in the INBOX.
> 
> 
> 
> 
> I have pretty much reached a dead end.  I am unable to determine the
> cause of this abnormal termination.  The logged messages don't give
> much help.  I can't tell if it is the primary server or test machine
> that is terminating ssh.  I have setup a second test machine.  Both of
> the test machines are raspberry pi 4s.  The real backup machine is
> intel.  On both the test machines, 2 of the three users are backedup
> properly with no errors.  Only one use has the issue.  It builds the
> directory structure correctly then starts transferring the INBOX cur
> files.  Eight files are transferred correctly before it stop.  It is
> the same set of 8 on both test machines.  That leads me to believe
> there is something funny in one of the messages, but which one.  There
> are over 100 messages in that directory.  The order of file transfer
> is not by date.  
> 
> 
> How do I get doveadm to tell me which file is being transferred when
> the problem occurs?  

if the problem is in one of the messages, perhaps change the test set.
for convenience, can temporarily rename/move things to get them out of
the way. divide test set in half, then in half again, etc., to zero in
on the one causing the issue, that'll show if a particular message is
problem or not.

once you find a problem message, if the why isn't obvious, i'd try
looking at it in hex, check permissions, etc.

j